Draft page for review before public site publication.
This is one pack in the Operator Agent Skills catalog: portable context packs and agent skills for Codex, Claude Code / cowork, OpenClaw, Hermes, and similar operator-agent runtimes.
Source series
Harness Engineering: The Runtime Layer Around AI Agents
Who it is for
Operators, founders, executives, managers, chiefs of staff, BizOps leads, product leaders, revenue leaders, and AI agents helping them make decisions from live operating context.
When to use it
Use this pack to convert harness engineering: the runtime layer around ai agents material into operator-grade decisions and reusable artifacts. The four skills are deliberately sequenced so diagnosis, mapping, planning, and review do not collapse into one vague prompt.
Included skills
- Agent Runtime Architecture Map — Use this workflow when you are working on Harness Engineering: The Runtime Layer Around AI Agents and need to make the system visible by mapping owners, flows, and evidence. Its distinct job inside the Harness Engineering Pack is to produce the runtime architecture map; use the neighboring skills for Eval Harness Spec, Context Window Management Plan, Runtime Failure Mode Review.
- Eval Harness Spec — Use this workflow when you are working on Harness Engineering: The Runtime Layer Around AI Agents and need to handle the system step without overlapping the other skills in the pack. Its distinct job inside the Harness Engineering Pack is to produce the eval harness spec; use the neighboring skills for Agent Runtime Architecture Map, Context Window Management Plan, Runtime Failure Mode Review.
- Context Window Management Plan — Use this workflow when you are working on Harness Engineering: The Runtime Layer Around AI Agents and need to turn the diagnosis into an executable plan with owners and gates. Its distinct job inside the Harness Engineering Pack is to produce the context management plan; use the neighboring skills for Agent Runtime Architecture Map, Eval Harness Spec, Runtime Failure Mode Review.
- Runtime Failure Mode Review — Use this workflow when you are working on Harness Engineering: The Runtime Layer Around AI Agents and need to diagnose the current state and name the operating risk. Its distinct job inside the Harness Engineering Pack is to produce the failure mode review; use the neighboring skills for Agent Runtime Architecture Map, Eval Harness Spec, Context Window Management Plan.
How to load in Codex / Claude Code / OpenClaw / Hermes
- Start from
START_HERE.md,manifest.json, or the main catalog. - Select this pack, then select one skill for the immediate job.
- Load only the pack
README.md, the selected skill file, and the matching example if a concrete prompt pattern helps. - Preserve the selected skill's output schema, assumptions, failure modes, and human review gates.
- Do not treat a skill as permission for autonomous external action; keep external, personnel, financial, legal, security, and high-impact decisions behind explicit human review.